Identification of metal-poor stars using the artificial neural network

Abstract

Context. Identification of metal-poor stars among field stars is extremely useful for studying the structure and evolution of the Galaxy and of external galaxies.

Aims. We search for metal-poor stars using the artificial neural network (ANN) and extend its usage to determine absolute magnitudes.

Methods. We have constructed a library of 167 medium-resolution stellar spectra (R ~ 1200) covering the stellar temperature range of 4200 to 8000 K, log g range of 0.5 to 5.0, and [Fe/H] range of −3.0 to +0.3 dex. This empirical spectral library was used to train ANNs, yielding an accuracy of 0.3 dex in [Fe/H] , 200 K in temperature, and 0.3 dex in log g. We found that the independent calibrations of near-solar metallicity stars and metal-poor stars decreases the errors in Teff and log g by nearly a factor of two.

Results. We calculated Teff, log g, and [Fe/H] on a consistent scale for a large number of field stars and candidate metal-poor stars. We extended the application of this method to the calibration of absolute magnitudes using nearby stars with well-estimated parallaxes. A better calibration accuracy for MV could be obtained by training separate ANNs for cool, warm, and metal-poor stars. The current accuracy of MV calibration is ±0.3 mag.

Conclusions. A list of newly identified metal-poor stars is presented. The MV calibration procedure developed here is reddening-independent and hence may serve as a powerful tool in studying galactic structure.